feat(charts): split 1× and 2× deployments by topology + show latest on /eval - #10
Merged
Merged
Conversation
…n /eval
Two fixes after the gemma4 ingest exposed gaps:
1. /inference series-key (Option A): getHardwareKey now appends a
"_{totalGpus}x" suffix when the deployment uses more than one GPU.
1× H100 and 2× H100 stop collapsing into a single "h100 (vllm, mtp)"
legend entry. Single-GPU runs stay un-suffixed for parity with
upstream legends. Disagg sums prefill + decode pools.
Same suffix added to buildAvailabilityHwKey via an optional
`totalGpus` parameter — callers that know the count (InferenceContext
from benchmark rows) can pass it; legacy callers (availability rows
lacking GPU counts) pass undefined and the suffix is skipped.
2. /evaluation no longer filters by selectedRunDate. The user reported
the "Batch size" filter chip not appearing on /evaluation: cause was
selectedRunDate being sticky at 2026-05-25 (set by an earlier
InferenceContext mount) while the mnbt rows are dated 2026-05-26.
The API correctly returned 12 of 19 rows, and categorizeTechniques
never saw a batch-size variant. RecipeComparison is a "current state
of recipes" view, not historical, so always use the no-date "latest
across all dates" path.
194/194 unit tests pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Two fixes after ingesting the gemma4 mnbt + spec-decoding runs.
1. /inference series-key — Option A (topology split)
`getHardwareKey` collapses 1× and 2× H100 into one `h100_vllm_mtp` series → one legend entry, one color, mixed dots. Fix: append `_{totalGpus}x` when totalGpus > 1.
Result on legend:
Disagg sums prefill + decode pools. Single-GPU runs keep the existing legend exactly.
`buildAvailabilityHwKey` gets the same optional `totalGpus` parameter.
2. /evaluation — drop sticky date filter
User reported "Batch size" chip missing on /evaluation. Root cause: `selectedRunDate` was sticky at `2026-05-25` from an earlier InferenceContext mount; the mnbt rows are dated `2026-05-26`, so the API filtered them out before `categorizeTechniques` ever saw them.
Fix: `RecipeComparison` is a "current recipe state" view, not historical. Always use the no-date / latest_benchmarks MV path. Empty-state copy simplified to "Pick a different model."
Test plan
🤖 Generated with Claude Code